首页 > 其他分享> 文章详细

【pytorch】深度学习

2022-06-30 22:02:23 阅读：202 来源： 互联网

标签：torch batch 学习 pytorch train cuda 深度 import model

【视频来源】

一 python安装环境

pycharm+anaconda

二创建python虚拟环境以及pip国内源设置

1创建：

目前兼容性比较好的版本

conda create -n py38 python=3.8

问题1：Collecting package metadata (current_repodata.json): faile 解决方案一

问题2：终端前有PS 输入CMD（PS表示pycharm用的终端是powershell.exe,如果需要转换为cmd.exe）

2激活

conda activate py38

>>>：命令交互符

CTRL+Z：回到命令行

3 设置pip国内源

（1）pip 的配置文件位置

linux/mac 系统
- 在用户目录下的 .pip/pip.conf 位置
windows 系统
- 在C盘用户目录下，建立 .pip/pip.ini 文件

（2）配置文件内容

阿里

[global]
trusted-host = mirrors.aliyun.com
index-url = https://mirrors.aliyun.com/pypi/simple

豆瓣

[global]
trusted-host = pypi.douban.com
index-url = https://pypi.douban.com/simple/

清华

[global]
trusted-host = pypi.tuna.tsinghua.edu.cn
index-url = https://pypi.tuna.tsinghua.edu.cn/simple

三 PyCharm基本用法及设置解释器

四实例1：手写字符识别

源代码：https://github.com/mivlab/AI_course

pip3 install torch torchvision torchaudio

pip install opencv-python

train_mnist.py

import torch
import math
import torch.nn as nn
from torch.autograd import Variable
from torchvision import transforms, models
import argparse
import os
from torch.utils.data import DataLoader

from dataloader import mnist_loader as ml
from models.cnn import Net
from toonnx import to_onnx


parser = argparse.ArgumentParser(description='PyTorch MNIST Example')
parser.add_argument('--datapath', required=True, help='data path')
parser.add_argument('--batch_size', type=int, default=256, help='training batch size')
parser.add_argument('--epochs', type=int, default=30, help='number of epochs to train')
parser.add_argument('--use_cuda', default=False, help='using CUDA for training')

args = parser.parse_args()
args.cuda = args.use_cuda and torch.cuda.is_available()
if args.cuda:
    torch.backends.cudnn.benchmark = True


def train():
    os.makedirs('./output', exist_ok=True)
    if True: #not os.path.exists('output/total.txt'):
        ml.image_list(args.datapath, 'output/total.txt')
        ml.shuffle_split('output/total.txt', 'output/train.txt', 'output/val.txt')

    train_data = ml.MyDataset(txt='output/train.txt', transform=transforms.ToTensor())
    val_data = ml.MyDataset(txt='output/val.txt', transform=transforms.ToTensor())
    train_loader = DataLoader(dataset=train_data, batch_size=args.batch_size, shuffle=True)
    val_loader = DataLoader(dataset=val_data, batch_size=args.batch_size)

    model = Net()
    #model = models.resnet18(num_classes=10)  # 调用内置模型
    #model.load_state_dict(torch.load('./output/params_10.pth'))
    #from torchsummary import summary
    #summary(model, (3, 28, 28))

    if args.cuda:
        print('training with cuda')
        model.cuda()
    optimizer = torch.optim.Adam(model.parameters(), lr=0.01, weight_decay=1e-3)
    scheduler = torch.optim.lr_scheduler.MultiStepLR(optimizer, [10, 20], 0.1)
    loss_func = nn.CrossEntropyLoss()

    for epoch in range(args.epochs):
        # training-----------------------------------
        model.train()
        train_loss = 0
        train_acc = 0
        for batch, (batch_x, batch_y) in enumerate(train_loader):
            if args.cuda:
                batch_x, batch_y = Variable(batch_x.cuda()), Variable(batch_y.cuda())
            else:
                batch_x, batch_y = Variable(batch_x), Variable(batch_y)
            out = model(batch_x)  # 256x3x28x28  out 256x10
            loss = loss_func(out, batch_y)
            train_loss += loss.item()
            pred = torch.max(out, 1)[1]
            train_correct = (pred == batch_y).sum()
            train_acc += train_correct.item()
            print('epoch: %2d/%d batch %3d/%d  Train Loss: %.3f, Acc: %.3f'
                  % (epoch + 1, args.epochs, batch, math.ceil(len(train_data) / args.batch_size),
                     loss.item(), train_correct.item() / len(batch_x)))

            optimizer.zero_grad()
            loss.backward()
            optimizer.step()
        scheduler.step()  # 更新learning rate
        print('Train Loss: %.6f, Acc: %.3f' % (train_loss / (math.ceil(len(train_data)/args.batch_size)),
                                               train_acc / (len(train_data))))

        # evaluation--------------------------------
        model.eval()
        eval_loss = 0
        eval_acc = 0
        for batch_x, batch_y in val_loader:
            if args.cuda:
                batch_x, batch_y = Variable(batch_x.cuda()), Variable(batch_y.cuda())
            else:
                batch_x, batch_y = Variable(batch_x), Variable(batch_y)

            out = model(batch_x)
            loss = loss_func(out, batch_y)
            eval_loss += loss.item()
            pred = torch.max(out, 1)[1]
            num_correct = (pred == batch_y).sum()
            eval_acc += num_correct.item()
        print('Val Loss: %.6f, Acc: %.3f' % (eval_loss / (math.ceil(len(val_data)/args.batch_size)),
                                             eval_acc / (len(val_data))))
        # save model --------------------------------
        if (epoch + 1) % 1 == 0:
            # torch.save(model, 'output/model_' + str(epoch+1) + '.pth')
            torch.save(model.state_dict(), 'output/params_' + str(epoch + 1) + '.pth')
            #to_onnx(model, 3, 28, 28, 'params.onnx')

if __name__ == '__main__':
    train()

test_mnist.py

import torch
import cv2
from torch.autograd import Variable
from torchvision import transforms
from models.cnn import Net
from toonnx import to_onnx

use_cuda = False
model = Net()
model.load_state_dict(torch.load('output/params_1.pth'))
# model = torch.load('output/model.pth')
model.eval()
if use_cuda and torch.cuda.is_available():
    model.cuda()

to_onnx(model, 3, 28, 28, 'output/params.onnx')

img = cv2.imread('4_00440.jpg')
img_tensor = transforms.ToTensor()(img)
img_tensor = img_tensor.unsqueeze(0)
if use_cuda and torch.cuda.is_available():
    prediction = model(Variable(img_tensor.cuda()))
else:
    prediction = model(Variable(img_tensor))
pred = torch.max(prediction, 1)[1]
print(pred)
cv2.imshow("image", img)
cv2.waitKey(0)

五实例2：手写字符识别

标签：torch,batch,学习,pytorch,train,cuda,深度,import,model
来源： https://www.cnblogs.com/chacaxx/p/16428754.html

本站声明： 1. iCode9 技术分享网（下文简称本站）提供的所有内容，仅供技术学习、探讨和分享；
2. 关于本站的所有留言、评论、转载及引用，纯属内容发起人的个人观点，与本站观点和立场无关；
3. 关于本站的所有言论和文字，纯属内容发起人的个人观点，与本站观点和立场无关；
4. 本站文章均是网友提供，不完全保证技术分享内容的完整性、准确性、时效性、风险性和版权归属；如您发现该文章侵犯了您的权益，可联系我们第一时间进行删除；
5. 本站为非盈利性的个人网站，所有内容不会用来进行牟利，也不会利用任何形式的广告来间接获益，纯粹是为了广大技术爱好者提供技术内容和技术思想的分享性交流网站。

ICode9